Comparison of Maximum Likelihood and GAN-based training of Real NVPs

Abstract

We train a generator by maximum likelihood and we also train the same generator architecture by Wasserstein GAN. We then compare the generated samples, exact log-probability densities and approximate Wasserstein distances. We show that an independent critic trained to approximate Wasserstein distance between the validation set and the generator distribution helps detect overfitting. Finally, we use ideas from the one-shot learning literature to develop a novel fast learning critic.

References in corpus (1)

• Towards Principled Methods for Training Generative Adversarial Networks

Cited by in corpus (8)

• Guided Image Generation with Conditional Invertible Neural Networks
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• The Cramer Distance as a Solution to Biased Wasserstein Gradients
• Variational Approaches for Auto-Encoding Generative Adversarial Networks
• Justifying Diagnosis Decisions by Deep Neural Networks
• Neural Density Estimation and Likelihood-free Inference
• A Method to Model Conditional Distributions with Normalizing Flows
• Generative Adversarial Data Programming